CIv15 – Autopoietic Planner via Compression‑Aligned Self‑Evolution

Hypothesis Statement

CIv15 evolves CIv14 into a self-maintaining, recursively optimizing system. Minimal generative programs are autonomously edited, future outcomes are simulated via decompression, and actions are selected to maximize compressibility and downstream utility. CIv15 operationalizes autopoietic planning in a measurable cybernetic framework.

---

Principles

1. Autopoiesis:

   • Program library L_t evolves to maintain or reduce MDL while improving forecast skill.
   • ΔMDL ≤ 0 over rolling window indicates internal viability.

2. Compression-Driven Planning:

   • Candidate actions are evaluated via decompression forecast.
   • Utility function U = f(ΔBDM, forecast accuracy, domain-specific reward).

3. Open-Ended Curriculum:

   • Sequence complexity progressively increases: climber → random → domain-transfer.
   • System’s φ-scored sketch output tracks adaptability.

4. Causal Robustness:

   • Perturb symbolic programs; successful recovery measured by return to φ ≥ φ_min and ΔMDL stabilization.

---

Mechanism

• Controller (Self-Editor & Planner):

  • Proposes edits to program library; evaluates via ΔMDL + φ + U.
  • Accepts edits that improve forecast, compressibility, or utility.

• Program Library:

  • Stores minimal generative programs with versioning.
  • Supports causal perturbation testing and counterfactual evaluation.

• Curriculum Engine:

  • Generates sequences of increasing complexity for continual learning.
  • Supports domain transfer tasks (e.g., ECA, ADIA sequences, market motifs).

• Simulator (Decompressor Forecaster):

  • Rolls forward program sketches to evaluate candidate actions.
  • Outputs utility-maximizing sequence/action pairs.

• Divergence Integration:

  • Latent-symbolic divergence + program evolution signal → adaptive regime probability.

• External Interface:

  • Provides actionable, φ-scored symbolic forecasts to downstream agents.

---

Success Metrics

• Autopoietic Viability: ΔMDL ≤ 0 with forecast MSE improving.
• Compression Efficiency: ΔBDM positive after self-edit cycles.
• Action Utility: Selected action maximizes U = f(ΔBDM, forecast accuracy, domain reward).
• Adaptability: Non-print φ outputs ≥ φ_min across curriculum levels.
• Robust Recovery: Successful post-perturbation recovery to stable ΔMDL/ΔBDM.

---

References

1. Hernández-Espinosa et al., 2024 – SuperARC
2. Zenil et al., 2018 – BDM/CTM complexity methods
3. Riedel & Zenil, 2025 – ECA rule minimality and causal decomposition
4. Maturana & Varela, 1980 – Autopoiesis
5. Ashby, 1956 – Design for a Brain
6. Burtsev et al., 2023 – Learning rules at the edge of chaos

---

Substrate Variants

Symbolic Substrate

• Mechanistic Role:

  • Represents minimal generative programs that describe the system’s behavior.
  • Supports self-editing and adaptation through the program library, enabling the system to modify its own structure in response to environmental changes.

• Key Signals:

  • φ (Program Length): Indicates the complexity of the symbolic representation; shorter programs are preferred for efficiency.
  • Δφ (Program Length Change): Tracks modifications in program length, reflecting adaptations to new regimes.
  • Edit Acceptance Metrics: Evaluates the success of proposed edits based on improvements in compressibility, utility, and forecast accuracy.

• Integration with Other Substrates:

  • Provides the foundational structure for the system’s behavior, guiding the generation of latent representations and informing forecasting models.
  • Interacts with the latent substrate to ensure that adaptations are consistent with observed data patterns.

Latent Substrate

• Mechanistic Role:

  • Encodes the statistical properties of sequences, capturing underlying patterns and dynamics.
  • Assesses the impact of symbolic edits on data compression and forecasting accuracy.

• Key Signals:

  • ΔBDM (Bayesian Description Length Change): Measures the effect of symbolic edits on the model’s posterior distribution.
  • ΔMDL (Minimum Description Length Change): Assesses changes in the model’s ability to compress data, indicating the efficiency of symbolic adaptations.
  • Forecast Error (MSE): Evaluates the accuracy of predictions post-edit, guiding further adaptations.

• Integration with Other Substrates:

  • Provides feedback on the effectiveness of symbolic edits, informing the program library’s evolution.
  • Collaborates with the symbolic substrate to ensure that adaptations lead to improved system performance.

Unified Substrate

• Mechanistic Role:

  • Orchestrates the integration of symbolic and latent representations to facilitate adaptive behavior.
  • Evaluates candidate actions based on decompression forecasts, selecting those that maximize utility and align with system goals.

• Key Signals:

  • Utility Function (U): Combines ΔBDM, ΔMDL, and forecast accuracy to evaluate the desirability of candidate actions.
  • Action Selection Metrics: Assesses the success of chosen actions based on their impact on system performance and adaptability.
  • Regime Adaptation Indicators: Monitors the system’s ability to adjust to new regimes, ensuring continued viability.

• Integration with Other Substrates:

  • Serves as the decision-making center, integrating inputs from symbolic and latent substrates to select optimal actions.
  • Ensures that adaptations are coherent and lead to improved system performance, maintaining internal consistency and external utility.

---